AgentFugue: Agent Scaling for Long-Horizon Tasks through Collective Reasoning

Researchers introduced AgentFugue, a multi-agent framework for long-horizon tasks.

Peer agents work on the same problem in parallel and share concise intermediate notes through a shared hub.

The hub records findings, attempts, and exclusions, enabling peer-to-peer collective reasoning.

Trained with supervised fine-tuning and end-to-end reinforcement learning, AgentFugue outperformed strong baselines in evaluation.